Baysian Theory of Mind: Modelling Human Mistakes
Humans have an innate ability to reason about the goals of others in their environment by observing their behavior. Remarkably, we, as humans, can do so even in the presence of mistakes or failures to achieve such goals. In this project, we explore the question of how to allow machines to have similar capabilities of reasoning about the mistakes of other agents in their environments. To do so, we model agents and their environments as generative processes that account for sub-optimality at three levels of decision-making: goal confusion with semantically similar goals, errors in planning, and mistakes in taking actions.
[Paper]